Context Engineer Workbench

Add background. Ask questions. Inspect the record.

Conversations are saved in this browser.

Panel 1

Injector

Add context

Paste the background information and instructions you want included with each question. Nothing is sent to the AI until you press Send.

About this context

This text is included once with every request. Website and GitHub links are not opened automatically.

Context SHA-256

…

A fingerprint of this draft text. It changes when the text changes. It does not measure answer quality.

0 bytes. Token counts, when available, appear in Diagnostics after you send.

Clear context empties the draft text. Saved conversations are not deleted.

What import and export include

Import loads text, context JSON, or a supported workbench export. Website loading and ZIP extraction are not supported.

Export context saves the current context name and text. Conversation messages are not included.

Delivery

Choose how the context is placed before your conversation.

System slot sends the context as a system message. User paste sends it as the first user message.

What delivery changes

These options change the context's role in the request. Neither retrains the model or guarantees that it follows every instruction.

Provider

Mock tries the workflow without calling an AI model. The result is a technical receipt, not an AI answer.

Grok requests an AI response through this app's server. It requires enabled access and may use paid API credits.

ChatGPT requests an AI response through this app's server. It uses its own server key, not the Grok key, and may use paid API credits.

Mock does not call a model.

Choose the provider yourself. A failed Grok or ChatGPT request will not be replaced with a Mock response.

Request settings

Start with the defaults. Adjust these settings when you need a different reply limit or are comparing request settings.

How long the app waits for input counting and the response. A timeout does not prove that the provider stopped processing the request.

Limit for the visible reply, not the full conversation. Increase it when an answer is cut short. A larger limit leaves less of the configured window available for input. Default is 512. Range is 16 to 8192. Status length means the reply was cut off here. Reasoning tokens are separate and are not part of this cap.

Controls randomness in text selection. Leave the default for your first conversation.

Controls how broad a set of likely text choices the model can sample from. Leave the default unless you are testing this setting.

Optional sampling seed. Leave blank for ordinary use. Setting a seed is not a guarantee of identical responses. Temperature 1 and top-p 1 are the provider defaults. Seed has no default.

Opens a conversation with no background text and no previous messages. Your other conversations remain saved.

Saved conversations

Reopen a conversation stored in this browser to read its messages and inspect individual turns.

Export all conversations saves every conversation this app holds in this browser, including context and turn records.